• No results found

The role of sleep in consolidation of multi item bound representations

N/A
N/A
Protected

Academic year: 2020

Share "The role of sleep in consolidation of multi item bound representations"

Copied!
283
0
0

Loading.... (view fulltext now)

Full text

(1)

The Role of Sleep in Consolidation of Multi-Item Bound

Representations

John James Shaw

MSc

This thesis is submitted in partial fulfilment of the requirements for

the degree of Doctor of Philosophy

Lancaster University

Department of Psychology

(2)

i

Prologue

This thesis has adopted the alternative format approach, thus chapters 2-5 have

been written in the style of journal articles suitable for submission. Each of these

chapters is independent and contains a review of the literature that is relevant to

the data presented as well as a discussion relating the findings to the wider

literature while Chapter 1 will provide a broader introduction to all of the work

undertaken in this thesis and identify the key research questions and aims.

In accordance with both department and university guidelines on alternative

format thesis submission each chapter includes a full bibliography as well as a

consolidated bibliography at the end of the thesis. At the beginning of each paper

there is a title page containing the paper title, its submission status, followed by

(3)

ii

Table of Contents

Prologue ……….. i

Table of Contents ………. ii

Declaration ………. iv

List of Tables ………. v

List of Figures ……… vi

Dissemination of Data ……… vii

Acknowledgement ……… ix

Dedication ……… xi

Thesis Abstract ……….. xii

1. Literature Review ……… 1

1.1 Overview .………. 2

1.2 Consolidation .……… 3

1.2.1 Origins of Consolidation theory ……… 3

1.2.2 Contemporary Theories of Consolidation ……….. 5

1.2.2.1 Complementary Learning Systems Theory ……… 5

1.2.2.2 Consolidation Through Reorganisation ……… 8

1.2.3 Conclusion ……… 9

1.3 Representations in Memory ………. 10

1.3.1 Formation of Conceptual Representations ……… 11

1.3.2

Role of Conceptual Representations in Supporting Episodic Memory 12 1.3.3 Consolidation of Bound Representations ……… 17

1.3.4 Maladaptive influence of Conceptual Knowledge in Consolidation …. 20 1.3.5 Conclusion ……… 23

1.4 Sleep’s Role in Memory Consolidation ……….. 24

1.4.1 Theories of Consolidation within Specific Sleep Stages ……… 26

1.4.1.1 Dual Process Theory ………. 26

1.4.1.2 Sequential Model of sleep stages ……… 27

1.4.1.3 Active System Consolidation Model ……… 28

1.4.2 Sleep Architecture ……….. 29

1.4.3 Sleep’s role in consolidation: Abstraction and Integration ………… 34

1.4.4 Role of Pre-existing Conceptual Knowledge in Sleep ……… 40

1.5 Overview and Objectives of the Thesis ……… 41 Chapter 2: The Role of Culture in Binding of Actions, Objects, and

Scenes in Visual Long-Term Memory

(4)

iii

Chapter 3: The role of sleep in Binding of Actions, Objects, and Scenes in Visual Long-Term Memory

44

Chapter 4: The Role of Sleep in Consolidation of Multi-Item Events and Differing Levels of Retrieval

45

Chapter 5: Lateralised Sleep Spindles Relate to False Memory Generation

46

Chapter 6. Thesis Discussion 47

6.1 Summary of Chapters ……….………. 47

6.2 Theoretical Contributions ……….……… 50

6.2.1 Differences in Consolidation by Item Modalities ……… 52

6.2.2 Differences in Specific Composite Recognition: An Action/Object Distinction 53 6.2.3 Role of Sleep in Consolidation of VLTM Based Bound Representations 58 6.2.4 The role of sleep in false memory generation ……….…… 61

6.2.5 Sleep Architecture in False memory generation ……….… 62

6.3 Methodological Limitations ……….……….. 65

6.4 Future Directions ……….……… 67

6.4.1 Role of Culture- Does a pictographic language help? ………... 67

6.4.2 Binding of Objects and Actions, why the gap? ……… 67

6.4.3 Role of schema in visual false memories ……….…………. 69

6.4.5 Replication of the nap study in an overnight paradigm ……… 69

6.5 Thesis Conclusion ……….……… 70

Consolidated Bibliography ……….……… 71 Appendices

(5)

iv

Declaration

The thesis contains original work completed solely by the author under the supervision of Professor Padraic Monaghan and Dr John Towse, and has not been submitted in the same form for the award of a higher degree at this institution or elsewhere.

Name: John Shaw

Signature:

Date: 19/05/2018

(6)

v

List of Tables Chapter 2.

Table 2.1. Table showing the total participants in each study by culture and ethnicity.

Table 2.2. Summary of the Linear Mixed Effect Model for the Action-Scene Recognition Task.

Table 2.3. Table showing mean proportion of objects recognised and average familiarity rating by culture.

Table 2.4. Table showing total participants in each study by culture and ethnicity. Table 2.5. Summary of the Linear Mixed Effect Model for the Action-Scene

Recognition Task.

Table 2.6. Table showing the average proportion of items correctly recognised for Western and East Asian cultures split by Task and foil type. SE in parentheses.

Chapter 4.

Table 4.1. Table displaying the average recall for each item modality by session. Recall defined as ratio of information recalled from session 1 in session 2.

Chapter 5.

Table 5.1. Proportion identified as ‘Seen-Old’ (SE in parentheses) for each word type, by hemisphere, and sleep or wake group. (LVF- Left Visual Field, RH- Right Hemisphere, RVF- Right Visual Field, LH- Left Hemisphere).

Table 5.2. Mean duration stage 1, stage 2, SWS, and REM sleep in minutes ± 1 SEM.

Table 5.3. Correlation coefficients between recognition accuracy, and proportion of total time slept in each sleep stage.

Table 5.4. Correlation coefficients between overall, LH and RH unseen lure false recognition with individual electrode sleep spindle density in sleep stage 2. Table 5.5. Correlation coefficients between seen-old and unseen-new recognition with lateralised sleep spindle density in sleep stage 2.

(7)

vi

List of Figures Chapter 1.

Figure 1.1. Standard model of Complementary Learning Systems. Figure 1.2. Dog smoking pipe.

Figure 1.3. Image depicting standard sleep stage activity over a single night.

Chapter 2.

Figure 2.1. Exp 1. Image demonstrating the study and test phase of the object-scene composite recognition task.

Figure 2.2. Exp 1. Object-scene Relational Binding: Graph showing the average accuracy for each culture by type of foil.

Figure 2.3. Exp 2. Image demonstrating the study and test phase of the action-scene composite recognition task.

Figure 2.4. Exp 2. Action-scene Relational Binding: Graph showing the average accuracy for each culture by type of foil.

Chapter 3.

Figure 3.1. Exp 1. Image demonstrating the study and test phase of the action-scene composite recognition task.

Figure 3.2. Exp 1. Sleep’s role in action-scene Relational Binding: Graph showing the average accuracy for each group by type of foil.

Figure 3.3. Exp 2. Sleep’s role in action-scene Relational Binding in Repeated

Viewing: Graph showing the average accuracy for each group by type of foil and graph showing the average accuracy for each experimental group compared across

Experiment 1a and experiment 2.

Figure 3.4 Exp 3. Image demonstrating the study and test phase of the object-scene composite recognition task.

Figure 3.5. Exp 3. Sleep’s role in object-scene relational binding: Graph showing the average accuracy for each experimental group.

Chapter 4.

Figure 4.1. Graph depicting mean number of details recalled for each Item modality split by group.

Figure 4.2. Graph depicting mean number of details recalled for each Item modality across sleep and wake groups.

Chapter 5.

Figure 5.1. Proportion of words endorsed as ‘Old’ by word type for sleep and wake groups by hemisphere.

(8)

vii

Dissemination of Data Contained Within Thesis Journal Publications

Shaw, J. J., & Monaghan, P. (2017). Lateralised sleep spindles relate to false memory generation. Neuropsychologia, 107, 60–67.

https://doi.org/10.1016/j.neuropsychologia.2017.11.002

Papers presented at scientific meetings

Shaw, J. J., & Monaghan, P. (January, 2017), The Impact of Sleep on the Binding of Actions, Objects and Scenes in Visual Long-Term Memory: Can repeated viewings help?, SARMAC XII, Sydney, Australia.

Shaw, J. J., & Monaghan, P. (August 2016), The Impact of Sleep on the Binding of Actions, Objects and Scenes in Visual Long-Term Memory, BPS

Cognitive Section Annual Conference, Barcelona, Spain

Shaw, J. J., & Monaghan, P. (July 2016), The Impact of Sleep on the Binding of Objects and Scenes in Visual Long-Term Memory. PsyPAG Annual

Conference, York, United Kingdom

Shaw, J. J., & Monaghan, P. (July 2016), The Impact of Sleep on the Binding of Actions and Scenes in Visual Long-Term Memory UK Sleep, Memory, and Language Meeting. Lancaster University, United Kingdom

Monaghan, P., Shaw, J. J., Ashworth-Lord, A., & Newbury, C. (January 2015), Lateralised False memories after a period of sleep. Royal Holloway University, United Kingdom

Posters presented at scientific meetings

Shaw, J. J., & Monaghan, P., (July 2017), Left Hemisphere Sleep Spindles Protect Against False Memories, Cardiff University Brain Research Imaging Centre Workshop, Cardiff, United Kingdom

Shaw, J. J., & Monaghan, P., (May 2017), Left Hemisphere Sleep Spindles Protect Against False Memories, Experimental Psychology Society Reading

(9)

viii

Shaw, J. J., & Monaghan, P. (January 2016), The Impact of Culture on the

(10)

ix

Acknowledgements

It is only as I sit down to write this section that I realise how many people have helped me though this thesis. First, I have to acknowledge and thank my supervisor, Padraic Monaghan. It is hard to believe that it has now been over 4.5 years since I first came to your office on the advice of Mark Howe to discuss a possible PhD project that merged my interest of false memories with sleep. I can honestly say I could not have dreamt for a better supervisor, from your gentle nudging of me to go beyond my comfort zone to what can only be described as your scary knowledge of near enough everything, thank you for everything that has gotten me to this stage. I would also like to acknowledge and thank John Towse, my second supervisor, for all the input in various meetings and annual reviews. Next, I must thank all the various members of Padraic’s lab group, of whom there are too many to name individually. Although the majority of you have no interest in sleep I always appreciated the comments and the outside perspective, it’s helped shape the thesis to what it is now.

(11)

x

where we can bamboozle them with sleep/false memory theory were worth the near-constant confusion!

A very special acknowledgement and thank you to Michelle To and Penny Lewis, my Viva examiners. Due to outside influences the time between submission and the Viva was only 17 days, including 5 of them being in a strike. I can’t thank you enough for powering through the thesis in that time. Despite all I was told before hand you both made the viva one of the most enjoyable experiences of the PhD. It was a complete joy to talk about my research and at no point did it feel like an ‘exam’, and for that I thank you.

Outside of Lancaster but remaining in academia I’d like to acknowledge the contribution of Scott Cairney for teaching me how to use polysomnography, and Justin Wood for very kindly sharing the action-scene/object-scene stimuli, without either of these people the thesis would have looked a lot different.

To Natalie, thank you for everything you have put up with during the thesis. From the long hours testing at 7am-10pm through to looking after me after I managed end up on crutches for 8 months from two separate incidents in the run up to submission, you have done so much for me I will never be able to pay back. You were more integral to the success of this thesis than you will ever be able to realise and that I will ever be able to show.

To my Mum, none of this would be possible without your encouragement and support throughout all my life. For all the times you have asked me how ‘College’ is going, or asking me when I will break up for the term (never), thank you. Your support throughout all of this has kept me going and knowing how proud you are of me is one of the most amazing feelings.

Finally, to my Dad. It is one of my biggest regrets that you can’t be here to share this achievement because it is as much a culmination of your

(12)

xi

Dedication

In memory of John Anthony Shaw (1946-2006)

(13)

xii

Abstract

A wealth of literature has demonstrated that in terms of memory

representations, items are not encoded equally. In part this has been suggested to be affected by conceptual knowledge, with items with pre-existing conceptual knowledge easier to encode and thus, more salient in subsequent representations. However, it is not known how this may affect different item modalities such as actions, objects, people, and locations. As such, this thesis seeks to understand the factors that can affect the different item modalities in multi-item memory representations primarily focusing on the role of sleep, which has been demonstrated to be beneficial for memory consolidation.

In paper 1 the role of culture (Western v East Asian) was examined in differences in bound memory representations, focusing on object-scene and scene pairings. It was observed that culture was not a factor for action-scene pairings but Western participants were significantly more accurate than East Asian participants for object-scene pairings. Furthermore, object-scene recognition was significantly more accurate than action-scene recognition across both cultures, suggesting a difference between actions and objects in ease of recognition.

In paper 2 this was expanded to see the role of sleep in consolidation of action-scene/object-scene pairings. As with paper 1 object-scene pairings were significantly more accurate than action-scene pairings in immediate testing, and object-scene recognition experienced a benefit of sleep, with accuracy higher in the sleep group than the wake group. This difference was not replicated in the action-scene task.

Paper 3 sought to apply the previous results to a practical situation, that of eyewitness testimony as the existing literature has revealed that within

(14)

xiii

with action details recalled to a higher level than object, location, and people details but there was no effect of sleep/wake.

Paper 4 moved away from visual memory and instead focused on the role of sleep on conceptual knowledge using the DRM paradigm. Participants were trained on 12 DRM wordlists and then were assigned either to a short-term wake group or a nap group monitored with polysomnography. At retrieval, sleep significantly increased lure-unseen acceptance compared to the equivalent period of wake, providing evidence of sleep promoting false memory generation.

Furthermore, RH spindle generation relative to LH spindle generation was

(15)

1

Chapter 1. Introduction

Chapter Summary

This chapter provides a review on the background literature that will be

explored in the thesis. Within the literature three key concepts will be

approached. First, the role of the hippocampus in consolidation of memory and

the theoretical perspectives within the current literature. Second, the theoretical

perspectives behind why memory distortions can occur, and third, the role of

sleep in memory consolidation. Finally, this chapter will provide an overview of

(16)

2

CH1. Literature Review

1.1 Overview

Imagine you are at Blackpool beach. You look out over the sea. The water has a murky-green colour as it mixes with the sand. Dark clouds are rolling in bringing rain. You notice a few seagulls and the differences between them. One has different colouring, is much smaller while the others are pristine white despite the water. One seagull suddenly swoops down and attacks a fish but stops once realising it is just a shoe. In our everyday life we see incredibly complex visual scenes with hundreds of details to encode ranging from the item-specific details (colour, size) to the whole image. Even more, there are several item modalities to process: the objects in the scene, the location, the actions performed, and the people present. There is a growing body of evidence for our understanding of how we can process these, with pre-existing conceptual knowledge playing an

important role and in a separate strand of research that item modalities are not equally consolidated either in quantity or fidelity of the representation. Recent literature suggests a role of the hippocampus in this form of processing (Horner & Doeller, 2017), yet one area of the literature that is relatively unexplored is the role of sleep in consolidation of these bound representations. Within the literature there is much evidence of the beneficial effect of sleep on consolidation of

information including both declarative and procedural (see Diekelmann & Born, 2010; Rasch & Born, 2013). More specifically, during sleep it is suggested that memories are reorganised from high-fidelity, highly-contextualised hippocampal-based episodic representations to a more conceptual, decontextualised

(17)

3

bound representations of differing item modalities and what factors may affect this. In the following section I will first address the various literature regarding consolidation theories then link that to research literature emphasising an

important role of both conceptual knowledge and hippocampal activity. After that I will discuss the role of sleep in consolidation and highlight where there is a gap in the literature that this thesis sets out to fill.

1.2 Consolidation

1.2.1 Origins of Consolidation Theory

The topic of memory consolidation is amongst the oldest within psychology with early studies on consolidating dating back to the late 1800s. By consolidation I refer to the established concept that after learning memory representations can be strengthened after they have been formed. In pioneering research, Ribot (1882) observed a number of patients reporting retrograde amnesia following a brain injury whereby they could recall information from their childhood but events temporally close to the event were unknown. From this Ribot proposed a temporal process of memory where memories become fixed as time progresses and

(18)

4

complete can act as interference to this consolidation, reducing subsequent recall. While the literature expanded in the early 20th century with studies seeking to replicate Müller and Pilzeckers’ (1900) results, it was fraught with issues. Buxton’s (1943) review highlighted a lack of standardisation across the field with many studies failing to account for confounds associated with repeated testing or type of stimuli used. As such, within cognitive psychology interest in consolidation gradually declined with contemporary theories of memory assuming an

unimportance of consolidation processes (see Brown & Lewandowsky, 2010 for a full review), instead preferring concepts such as interference or decay (Wixted, 2004, 2005) to account for difference in memory retention over time.

(19)

5

store that is unable to be related to existing memories and in order to overcome this, these simple memory representations are transferred from the simple memory store (that Marr suggested to be within the archicortex), to the neocortex. Of particular relevance to the thesis, Marr emphasised this transfer occurs mostly during sleep.

1.2.2 Contemporary theories of Consolidation

The work of Hebb (1949) and Marr (1970, 1971) has since been used as the foundation for modern consolidation theories, resulting in theories containing a strong biological and cognitive framework. Within the literature supporting consolidation they can be split into three key models, The Complementary Learning Systems model (CLS; McClelland et al., 1995; Squire, 1992), the Multiple Trace Theory (MMT; Nadel, Samsonovich, Ryan, & Moscovitch, 2000), and the Transformation Model of Consolidation (TMC; Winocur et al., 2010). While each model may differ in specific processes, there is a consensus across all

models regarding two key aspects. 1) that memory can be split into two forms of representation- episodic and semantic/conceptual (hereafter termed conceptual representation), 2) that the hippocampus and the neocortex are integral to these representations. In the following sections I will first describe and review each model of consolidation, highlighting the strengths and weaknesses in contrast to the other models and then discuss the different representations.

1.2.2.1 Complementary Learning Systems Theory

Building upon the word of Hebb (1949) and Marr (1970, 1971), the Complementary Learning Systems model of consolidation is perhaps the most widely known and researched amongst the literature (Squire, 1992; McClelland et al., 1995). Within the CLS model the hippocampus encodes sparse,

(20)

6

highly-overlapping, decontextualised representations that are susceptible to change over a period of time (conceptual), and through an interleaving of experiences, avoids the issue of catastrophic interference at reactivation:

neocortical representations are long lasting while hippocampal representations are temporary. Within the model the neocortical representations are often initially represented quite poorly, with only weak connections existing between

representations (Murre, 1996; Meeter & Murre, 2004a, 2004b), suggested to be in geographically disparate areas (Alvarez & Squire, 1994), and are relatively “slow”. This key characteristic of it being the slow learning system is suggested to be essential to long term consolidation, with McClelland et al. (1995) emphasising the importance of this store as essential to ensure that prior knowledge is not overwritten by incoming information thus impairing subsequent retrieval, a phenomenon termed catastrophic interference and previously demonstrated in artificial intelligence (Ratcliff, 1990). As such, within the CLS model consolidation occurs as repeated hippocampal reactivation reinstates the disparate neocortical associations further strengthening them until the point where the episodic memory within the neocortical representation is a replication of the original hippocampal representation and as such, the latter is no longer required and gradually decays (Figure 1.1).

[image:20.595.117.501.531.651.2]

Time

Figure 1.1. Standard model of Complementary Learning Systems. The model assumes two distinct memory systems, a hippocampal-based fast learning store and a neocortical-based slow learning stores, with both occurring in parallel. Over time through repeated activation the hippocampal store seeks to strengthen the neocortical connections until the hippocampal representation is no longer required. Figure adapted from Frankland & Bontempi (2005).

Neocortical=based slow learning store

(21)

7

Within the literature there is substantial support for the CLS model, with evidence in both human and animal studies. Within human studies the model is demonstrated mainly in studies on amnesic patients with hippocampal or parahippocampal lesions, primarily through hypoxic patients (loss of oxygen resulting in neuronal damage), of which the hippocampus is particularly

(22)

8

with hippocampal-neocortical connectivity also decreasing by the second testing, supporting the suggestion that hippocampal representations are not as essential over time.

1.2.2.2 Consolidation Through Reorganisation

However, there is a growing body of evidence that contradicts key points of the CLS model. Since patient HM several studies have observed no temporally graded retrograde amnesia, suggesting no memory impairment, or a complete memory impairment, in some cases lasting decades (Bayley et al., 2006; Manns et al., 2003; Nadel, Winocur, Ryan, & Moscovitch, 2007), a result not compatible with CLS. As such, alternative models of consolidation have since been promoted. Nadel et al.’s (2000) Multiple Trace Theory (MMT) suggests differing roles for the two systems similar to the standard consolidation theories (SCT), memories are initially encoded in both the hippocampal and neocortical, with the neocortical representation considered to be highly-semantic and context-independent while the hippocampal representation is suggested to be a high-fidelity, contextualised representation of an episodic memory. Repeated reactivation of the hippocampal representation leads to multiple distinct traces created for the episodic memory, ensuring a long-lasting representation while also promoting common information for integration within the neocortical semantic network. Once the semantic

representation is strong enough, these decontextualised memory traces no longer require hippocampal activation to be retrieved but the specific episodic memory still required the hippocampal-cortical connection.

Evidence for the MMT can be seen in studies of the dissociation between episodic and semantic memory. For example, there are reports of amnesic patients that show temporally ungraded impairment for episodic memory

(23)

9

‘thatcherism’, demonstrating that retrograde amnesia did not impair semantic memory, suggesting a dissociation between the episodic and semantic memory. Using fMRI, Davachi, Mitchell, and Wagner (2003) observed hippocampal and perirhinal activity in participants at both encoding and retrieval within an item recognition task and observed that encoding activation in the hippocampus and posterior parahippocampal cortex predicted later item recognition.

More recently, in an effort to unify CLS and MMT a review by Winocur et al. (2010) summarised the literature by proposing the Transformation Model of Consolidation, encompassing key themes of both CLS and MMT. The key aspect of the transformation model is that the neocortical representation is significantly different to that of the hippocampus by virtue of their complementary nature. The hippocampus encodes a high-fidelity, contextualised episodic memory whereas the neocortex encodes a conceptual, decontextualised schematic memory and it is through repeated reactivation that the schema is modified and formed. The

schematic structure allows key commonalities between memories to be consolidated in the long-term memory, allowing retrieval of the schematic

representation when relevant information is experienced. Furthermore, Winocur et al. emphasise that the hippocampal representation is not as transient as CLS models would suggest, instead the original episodic memory remains in the hippocampus for as long as it can be retained. In doing so the model is able to account for the variation in retrograde amnesia observed in the literature.

1.2.3 Conclusion

(24)

10

interaction between these representations. The CLS model adopts a relatively unidirectional view that the hippocampal episodic representation serves to

consolidate the neocortical conceptual representation and once this has occurred, is free to degrade as the remote memory is available within the neocortical store. In contrast, MMT and TMC (Nadel et al., 2000; Winocur et al., 2010) adopt a view of greater hippocampal involvement in long-term memory, with the hippocampal-episodic representation serving to influence the conceptual representation but is essential for providing contextual, spatial, and temporal detail that is lost in the decontextualised, conceptual representation. Moreover, there is a growing literature suggesting that conceptual representations are also important for encoding of episodic representations. In the following sections I will first describe the literature surrounding conceptual representations, including what they are and how they are formed, before linking them to episodic memory retrieval.

1.3 Representations in Memory

Early models of conceptual representations suggested that they were stored in separate modality-specific representations of motor, sensory, and verbal

information, and that conceptual knowledge represented a form of network between these modality-specific representations within a distributed neocortical network (Eggert, 1977; Martin, 2007; Patterson et al., 2007). However,

subsequent evidence of semantic dementia appears to dispute this distributed view as damage to the temporal lobe results in a breakdown of conceptual knowledge across all domains, providing contradictory evidence of a distributed model of representations (Bozeat et al., 2000; Lambon Ralph et al., 1999, 2001; Piwnica-Worms et al., 2010).

(25)

11

et al., 2004), with the spokes forming associations through the hub. The purpose of the hub is thought to act in three key ways: integration of information across different modalities and experiences; encoding the structure of conceptual relationships via identification of commonalities across items in order to generalise to novel items; and to encode items in a way in order to allow high-order generalisation in high-order to allow generalisation across items that may not necessarily have the same surface-level features (e.g., knowing that a dolphin is a mammal despite all visual features indicating it may be a fish) (Rogers et al., 2004; Lambon Ralph et al., 2010; Lambon Ralph, 2014). Evidence supporting a transmodal hub model was recently demonstrated in a computational model by McClelland and Rogers (McClelland & Rogers, 2003; Rogers et al., 2004; Rogers & McClelland, 2004). Within their model modality-specific regions require additional support from a central representational unit (the hub) in order to allow cross-modal mappings and encoding of conceptual relationships. Studies on patients with semantic dementia suggest the hub is located within the anterior temporal lobe (ATL), with atrophy of the ATL linked to specific breakdown of conceptual knowledge (Brambati et al., 2009; Hodges et al., 1992; Mion et al., 2010), with a positive correlation observed between the degree of damage to the ATL and the severity of semantic impairment (Mummery et al., 2000; Nestor et al., 2006). This, combined with its close proximity to the medial temporal lobe (MTL), limbic system, and frontal cortex allow a high level of interconnectivity across important regions (Patterson et al., 2007).

1.3.1 Formation of Conceptual Representations

But how are conceptual representations formed? much of the evidence for conceptual representations suggest a form of long-term learning of elements through repeated exposure or reactivation of hippocampal based episodic representations. For example, within the previously mentioned CLS model

(26)

12

fast-learning episodic representation in the hippocampus and a slow-learning conceptual representation in the neocortex. Through repeated hippocampal activation of the episodic representation, the commonalities across experiences and exemplars are extracted and form a conceptual representation independent of the episodic representation, a process referred to as semanticisation (Battaglia et al., 2011; Cermak, 1984; McClelland et al., 1995; Moscovitch et al., 2005; Meeter & Murre, 2004; Rosenbaum et al., 2001; Stickgold, 2009; Sweegers & Talamini, 2014; Westmacott et al., 2004).Within the literature there appears to be two distinct forms of conceptual formation, that of explicit, verbalised rule based learning and implicit, complex, non-verbal learning (Ashby et al., 1998; Knowlton & Squire, 1993; Nosofsky & Johansen, 2000). For the former, there are numerous studies that demonstrate the use of simple, verbalised rule based learning in various classification tasks whereby participants are able to explicitly define the categorisation rule (Ashby & Spiering, 2004; Ashby & Maddox, 2005). In complex learning, one suggestion for conceptual formation is that of similarity based-models (Ashby & Waldron, 1999). In similarity based based-models, categorisation of items is decided based upon similarity to existing conceptual representations (Ashby & Maddox, 2005; Medin & Rips, 2005; Palmeri & Flanery, 1999; Posner & Keele, 1968; Reed, 1972; Smith, 2001; Smith & Minda, 2002; Tunney & Fernie, 2012). An extension of this theory is that of an exemplar-based model of

semanticisation (otherwise termed episodic models; Smith et al., 1998; Tunney & Fernie, 2012). Within this theory conceptual representations are formed through a composition of previously encountered category exemplars and activation of episodic representations allows subsequent categorisation of a novel item.

1.3.2 Role of Conceptual Representations in Supporting Episodic Memory

(27)

13

and hippocampal representations, much of this is theorised to be unidirectional whereby the hippocampal-episodic representation forms the basis of the

neocortical-conceptual representation, with little detail given to the role of conceptual knowledge in the hippocampus. This appears somewhat at odds with neurobiological literature, with neurobiological models of episodic memory such as the Hierarchical Relational Binding Theory (hRBT) of Shimamura (2010)

demonstrating that the hippocampus is key to forming associations between multiple memory representations, acting as a “convergence zone” whereby disparate inputs from the MTL and neocortex are fed and bound into a single representation (Damasio, 1989; Marr, 1971; McClelland et al., 1995) that ultimately supports episodic memory, requiring both episodic and conceptual knowledge in order to create an episodic representation (Tulving, 1983). More specifically, within the hippocampus specific concept cells have been observed in relation to conceptual knowledge, such as animals, buildings, and celebrities (Quiroga, Reddy, Kreiman, Koch, & Fried, 2005; Kreiman, Koch, & Fried, 2000). For example, it was observed that firing of concept cells is not linked to a

particular trigger, but rather their firing pattern could be activated through multiple associated stimuli, such as a written form of a celebrity’s name or a photo (Quiroga et al., 2005). Moreover, concept cells can tune their firing field rapidly, with rodent studies demonstrating that place cells can tune after a single visit to the target location (Monaco, Rao, Roth, & Knierim, 2014), and concept cells can respond to a person within hours of meeting them (Quiroga, Kraskov, Koch, & Fried, 2009). Based upon this Quiroga (2012) has suggested that these place and concept cells form the ‘building blocks’ of hippocampal bound

representations acting as a form of early precursor of conceptual knowledge, allowing rapid encoding and formation of high-fidelity hippocampal

(28)

14

and computational modelling studies suggest that pre-existing conceptual

knowledge can be advantageous to consolidation of novel but related material. Within rodents, Tse et al. (2007) trained rats within a location on a schematic food layout, allowing learning of the typical locations via smell. Rats were given hippocampal or control lesions at different points in training with neocortical retrieval measured by digging at previously learned food sites. While the rats lesioned 3h after learning could not generalise to novel-but-consistent reward probes, rats lesioned after 48h were able to generalise to schema consistent novelty, suggesting a time-dependent measure in consolidation of newly learned information. Furthermore, when reward probes were delivered in a schema-inconsistent manner, both task improvement and generalisation was impaired for both sets of rats. The study demonstrated that information that is compatible with pre-existing schemas can aid consolidation, even when hippocampal lesions were implemented after learning of the initial schema. Within humans, van Kesteren et al. (2010, 2013, 2014) observed through fMRI studies that prior knowledge modulated the contribution of the neocortex and hippocampus to encoding of new information.

But for the purpose of the thesis I wish to focus the beneficial effect of conceptual knowledge on visual long-term memory (VLTM). At an item level it has been demonstrated that conceptual knowledge of visual elements can affect subsequent retrieval (Esyenck, 1979; Nairne, 2006; Rawson & Van Overschelde, 2008; Schmidt, 1985; von Restorff, 1933). In a series of experiments Koutstaal et al. (2003) observed that recognition of real-world objects was better than

memory for highly-distinct but novel items, suggesting that even with explicit labelling of ambiguous shapes, there is a substantial benefit of pre-existing conceptual knowledge in recognition of shapes. Furthermore, there is recent evidence that even within VLTM, the level of detail in the conceptual

(29)

15

(2010a, 2010b) sought to compare the role of categories for various visual items in VLTM. Extending the previous literature that demonstrates that humans can have a large a detailed storage capacity to the extent of maintaining thousands of high-fidelity representations (Standing, 1973; Brady, Konkle, Alvarez, & Oliva, 2008; Hollingworth, 2004; Vogt & Magnussen, 2007), Konkle et al. examined the role of categorical and perceptual features supported long-term memory

representations. Within their first study (2010a) participants viewed 2800 objects with a varying number of exemplars from each category and in a subsequent two alternative forced-choice task (2AFC) participants were presented with one of the previously seen exemplars and a second, unseen exemplar from the same

category. Accuracy was consistently high across all exemplar levels, even with 16 presented exemplars (82%) but memory performance decreased as more

exemplars were displayed, suggesting that while VLTM is capable of storing a large number of category exemplars, it is prone to interference when multiple items are required to be encoded. Furthermore, variations in categorical distinctiveness (e.g., a category of car could vary from a mini through to a limousine) were also found to reduce interference suggesting that while the conceptual representation can support recognition of an item, key contextual details of the item are important for specific recognition (e.g., although a limo is a car, knowing the key detail of it being longer than the average car). In a follow up study Konkle et al. (2010b) demonstrated a similar recognition rate for scene memory. Once again participants were presented with over 2900 images of scenery from 128 categories with 1, 4, 16, or 64 exemplars displayed per category. In a subsequent 2AFC task recognition rate was still consistently high, with the novel condition reporting 96% recognition accuracy against a novel item from an unseen category, dropping to 76% accuracy in the 64-exemplar category, comparable to the results of the object recognition task (Konkle et al., 2010a).

(30)

16

as such, items within the VLTM must be represented at a conceptual level through an item prototype, a form mental representation of an item that can be linked to multiple exemplars in order to reduce cognitive load and aid consolidation of an episodic representation through accelerated consolidation of key features to be used for subsequent recognition. For example, when presented with a limousine the item prototype can automatically encode the key features of a car (four wheels, general shape) while key details can be formed in the episodic

representation (in the example of a limo, the extended length of the car). In a subsequent recognition task, if participants are presented with the limo and another car, the prototype of a car can feature multiple exemplars ranging from standard hatchbacks, SUVs, cabriolets, even motorised wheelchairs, but are all able to be identified as a car but the specific episodic/contextual details of it being a limo can aid consolidation. The result is akin to examples form the verbal domain whereby familiar letter strings (e.g., FBI-BBC-NHS-MAN-UTD) are more likely to be retrieved than unfamiliar letter strings (FBIB-BCBH-SMA-NUTD; Bower &

(31)

17

1.3.3 Consolidation of Bound Representations

As of yet all the discussed studies have primarily focused on a single-item presentation and how episodic and conceptual representations may form from there, but real life is rarely so singular, visual scenes are incredibly complex and include a variety of different items. For example in the image below (Figure 1.2), although the primary focus is the dog, there are multiple different items of varying complexity, including the dog (breed, colour, spatial orientation, size), the pipe (colour, size, style), and the table (colour, shape, size). As such in order to create an episodic memory of those specific elements, consolidation of all the items and associations between these must be formed early in encoding.

[image:31.595.144.396.310.639.2]

Figure 1.2. A dog smoking a pipe. Taken from stimuli from Standing (1973).

(32)

18

zone’ through three key mechanisms: 1) rapid learning of associations, 2) pattern separation, 3) pattern completion.

Rapid learning of new associations

As it is incredibly unlikely that you will often experience the exact same scenario multiple times, one of the functions of the hippocampus is suggested to be rapid learning of new associations that are already encoded as conceptual knowledge. Several theories posit that the hippocampus is central in this form of rapid learned association through its position within the MTL, acting as a “hub” for separate elements. Recent studies utilising fMRI support the concept of bound representations being represented in the hippocampus yet these mostly focused on the hippocampus in isolation, neglecting the wider neocortical network. One study that built upon these findings was that of Backus et al. (2016) who examined hippocampal activity using an associative memory task then measuring

hippocampal activity through multivariate pattern analyses of fMRI data. Backus and colleagues observed overlapping coding within the hippocampus alongside a hub-like network, supporting the idea that information from various neocortical sites feed into the hippocampus and that this is crucial for bound representation formation.

Pattern Separation

(33)

19

(e.g., two similar rubber ducks) and noted that measures of hippocampal activity decreased over repeated presentation of the stimuli (Bakker, Kirwan, Miller, & Stark, 2008), which Bakker and colleagues inferred to be the presence of pattern separation.

The effect of pattern separation can also extend to complex events with overlapping content. In a study using video stimuli Chadwick, Hassabis, and Maguire (2011) presented participants with multiple videos that combined two background and two foreground stimuli in unique pairings (e.g., stimuli a, b, c, d, could be combined as ac, ad, bc, bd). Using fMRI Chadwick and colleagues

observed that the hippocampal representations could be distinguished within each video, suggesting that even in input that is highly overlapping the hippocampus is capable of producing high-fidelity, separated representations.

Pattern Completion

(34)

20

(e.g., visual differences between species of banana), then it may be that

conceptual knowledge of the overall category may be preferentially retrieved, reducing fidelity.

1.3.4 Maladaptive Influence of Conceptual Knowledge in Consolidation

However, although prior conceptual knowledge can be of great benefit for expedited encoding and consolation of schema-consistent information, this same strength can also be maladaptive in situations whereby the information is

inconsistent with our prior knowledge. Early work by Bartlett (1932)

demonstrated an early representation of a conceptual consolidation within the “War of the Ghosts”. Within the study participants read an unfamiliar story that

contained culturally unfamiliar narratives set within Native American folklore but contained information not consistent with the setting. Over time participants began to make errors of commission in which they inserted unrelated information into the narrative as well as errors of omission, failing to produce information that was inconsistent with the theme, with false memories an error of commission. Importantly, Bartlett noted how the information provided by participants over time gradually changed to be more consistent with pre-existing views on Native American Folklore, with errors of omission reducing inconsistent information, and errors of commission introducing consistent. Bartlett’s study was among the first in the literature to demonstrate that consolidated memory is not a direct

representation of what was experienced, rather an interpretation. Yet despite the implications of Bartlett’s study, the results have never been successfully

replicated by researchers (Gauld & Stephenson, 1967; Roediger, Wheeler, & Rajaram, 1993). In fact, Wheeler and Roediger (1992) observed that after presenting participants with the original ‘War of the Ghosts’ story memory was improved over repeated tests, but only if short delays occurred between the initial

(35)

21

memory, and Reconstructive Memory, emphasizing the process of filling in

missing elements, typically with information that is contextually correct, with these distinctions serving as precursors to the episodic and semantic (conceptual) memory outlined in subsequent models (e.g., the CLS model, Squire 1992;

McClelland et al., 1995).

Further literature on flashbulb memories has demonstrated that memories for significant world events such as the terrorist attack of September 11th, 2001 are typically reported with high confidence and in high fidelity but are subject to change even within 1 year of the original event and can distort even further over time (Brown & Kulik, 1977; Neisser, 1982, 1986; Lee & Brown, 2003; Conway, Skitka, Hemmerich, & Kershaw, 2009). For example, 50% of participants surveyed in the years after September 11th Terrorist attack report seeing the first plane hit the tower, including then U.S President George W Bush who confidently reported seeing the first plane hit the building, despite the event not being televised

(36)

22

maladaptive influence of conceptual knowledge, shifting the episodic

representation in order to fit in with an expected schema.

In the past two decades the most prominent form of schema-based false memory generation has come in the form of the Deese-Roediger-McDermott (DRM) paradigm (Deese, 1959; Roediger & McDermott, 1995). In Deese’s original study, one overlooked until Roediger and McDermott adopted its paradigm,

participants viewed a word list then completed a free recall memory task. The primary purpose of the study was to look at extra list intrusions in the free recall. Participants viewed 36 word lists containing 12 words each and each one had a critical associate. For example, the critical word King had a word list containing queen, England, crown, Prince, George, dictator, palace, throne, chess, rule,

subjects, monarch, royal, leader & reign. Deese observed that some lists reliably

induced the critical word on immediate free recall. The result of Deese’s study was replicated by Roediger and McDermott using six of the critical words from Deese’s study. In their study when participants completed a free recall task recall of critical lures (e.g., for the above example, King) was extremely high at 55% recall, a result that was higher than words presented in the middle of the study phase and a replication of Deese’s original study. When participants were further questioned about the false recall participants regularly responded as having ‘known’ that the critical word appeared on the original word list. A follow up study by McDermott & Roediger (1998) found that when participants were explicitly instructed to decide carefully whether the critical word had previously appeared the false recall phenomenon still existed. Subsequent research by Neuschatz, Benoit, and Payne (2003) found that only when participants were given an explicit warning on the false memory effect was the effect reduced.

Since its inception the DRM paradigm has been used extensively using both visually and aurally presented word lists to explore various effects, including long-lived semantic priming (McKone & Murphy, 2000), memory illusions

(37)

23

Roediger, 2003). The effect has also been expanded beyond a word list with false memory effects appearing in picture stimuli as well (Roediger, McDermott, & Robinson, 1998; Gallo, McDermott, Percer, & Roediger, 2001; Lutz et al., 2017; Roediger, McDermott, Pisoni, & Gallo, 2004). One recent study, that of Lew and Howe (2016) presented participants with images that contained items that were either schema-consistent (e.g., frying pan on the cooker), schema-inconsistent (e.g., toaster under the table), or schema-irrelevant (e.g., toilet brush on the cooker). In a subsequent recognition task, participants were highly accurate in schema-consistent item locations and schema-irrelevant items, but they observed that schema-inconsistent items were more likely to be falsely recognised as in their schema-consistent locations (e.g., toaster on the worktop), with Lew and Howe theorising that this was in part due to a top-down schema influence on encoding of the episodic memory to make it more consistent with the pre-existing conceptual knowledge. Although it is to my own knowledge the first to explore false memory generation in visual memory in this form, the result is consistent with literature from visual memory. Although it is only recently published the study demonstrates that schema-influence is not limited to just a semantic domain but also visual. Given that these false memory generation effects can occur within an hour of encoding, this suggests that conceptual knowledge has a strong influence on encoding and retrieval of declarative memory, possibly as a result of the expedited consolidation brought about by the items all being consistent within a specific conceptual schema.

1.3.5 Conclusion

(38)

24

of a beneficial role of prior knowledge in consolidation of novel information,

suggesting that conceptual representations can also affect episodic

representations in a form of top-down processing, with conceptual knowledge supporting both encoding through accelerated encoding and retrieval of

information through a form of pattern completion. Importantly, it appears that the hippocampus is essential for both initial formation of conceptual knowledge and integration of conceptual representations and episodic, with the hippocampus acting as a form of convergence zone for both new and pre-existing information.

However, there is an unresolved issue: Although there is evidence that pre-existing knowledge can aid specific episodic retrieval, it is unknown what the effect of sleep may have. While recent studies have demonstrated a beneficial effect of sleep on integration of information related to pre-existing knowledge (Hennies et al., 2016), It is unknown what the relation is to episodic representations, similar to that of Konkle et al. (2010a, 2010b). In the following sections I will discuss the role of sleep in regard to the wider literature before focusing more on the role of sleep in abstraction and integration of information.

1.4 Sleep’s Role in Memory Consolidation

As previously stated in the introduction to consolidation, the role of sleep in consolidation has been long known, with Marr (1970, 1971) and McClelland et al. (1995) suggesting hippocampal activity during sleep aided memory

(39)

25

procedural tasks (for a summary, see. Diekelmann & Born, 2010; Marshall & Born, 2007; Rasch & Born, 2013). Physiologically, sleep can be divided into stages, rapid eye movement (REM) and non-REM (NREM), which can be further broken down into stage 1 (S1), stage 2 (S2), and SWS (stage 3 and stage 4 combined), with SWS occurring more than REM at the start of the night while this pattern reverses as the night goes on (Figure 1.3). Within the literature there are multiple theories as to sleep’s role in consolidation, those that are stage

dependent, such as the dual process theory and the sequential hypothesis, and

those that are more physiologically based, such as the active system

consolidation model. Although they are not mutually exclusive, they assume different processes, the former focusing more on the entire sleep stage while the latter is more focused on the neurobiological events within each stage. In the following sections I will discuss these theories, including their strengths,

[image:39.595.108.530.483.673.2]

weaknesses, and relevance to the thesis. After discussing the theories I will focus on the role of specific sleep architecture and finally examine the literature on the role of sleep in abstraction and integration of information.

(40)

26

1.4.1 Theories of Consolidation within Specific Sleep Stages

1.4.1.1 Dual Process Theory (DPT)

Within the dual process theory of sleep, each specific sleep stage is

thought to be beneficial for a memory type, SWS is thought to benefit declarative memory while REM sleep is beneficial for nondeclarative memory. This theory is primarily supported by studies employing the “night-half” paradigm developed by Ekstrand and Colleagues (Yaroush, Sullivan, & Ekstrand, 1971; Fowler, Sullivan, & Ekstrand, 1973; Ekstrand, 1972). Within the paradigm performance is compared across sleep intervals of either the first half or the late half of nocturnal sleep, taking advantage of the SWS dominant first half of sleep and the REM dominated second half to allow comparisons without the confound of repeated awakening. Yaroush et al. (1971) demonstrated the benefit of SWS early sleep compared to REM late sleep in a study on associative word pairs while a study by Barrett and Ekstrand (1972) replicated the effect but controlled for circadian rhythms by manipulating sleep so that both “night-halfs” occurred in the same circadian phase. Plihal and Born (1997, 1999) explored the difference between the SWS-rich early sleep and the REM-SWS-rich late sleep further, training participants

separately on a word pair task and a separate mirror tracing task. They observed that SWS early sleep was beneficial to the word pairs task while REM late sleep aided the mirror tracing task, supporting the dual process theory (DPT).

(41)

27

Helms, & Born, 2002; Genzel, Dresler, Wehrle, Grozinger, & Steiger, 2009;

Walker, Brakefield, Morgan, Hobson, & Stickgold, 2002; Ruch et al., 2012, van der Helm et al., 2011). For example, sleep after learning a new procedural motor task has been demonstrated to not just increase the time spent in S2, but also the spindle density in S2 (Fogel & Smith, 2006; Fogel, Smith, & Cote, 2007; Nishida & Walker, 2007), with both of these increases correlated with subsequent increased performance in the motor task from pre- to post-sleep. More recently, van der Helm, Gujar, Nishida, and Walker (2011) investigated the impact of sleep on an episodic memory task using an item and context learning paradigm. Within the study participants learned two word lists, each associated with a different context and then assigned to either a nap or no nap group. In a subsequent recognition task they observed that for item memory there was no beneficial effect of sleep, but for context memory, which is hippocampal dependent, there was a significant benefit of sleep, with accuracy significantly correlating with amount of S2 sleep, a result supported by Ruch et al. (2012) who demonstrated S2’s role in consolidation of declarative memories. In Ruch et al.’s (2012) study, participants learned associations between faces and cities and after a nap

completed a cued recall task, with sleep stages and spindle activity monitored during sleep. Increases in spindle activity in S2 were associated with enhanced retention of face-city associations.

1.4.1.2 Sequential Model of Sleep

(42)

28

participants completed the full sleep cycle, going through SWS and REM multiple times performance increased significantly, strongly supporting the concept that sequential sleep may be more beneficial than any one stage. Elsewhere, retention of verbal material has been correlated with both early SWS and late REM, again supporting the view that repeated cycles are more important than any one stage (Ficca, Lombardo, Rossi, & Salzarulo, 2000; Mazzoni et al., 1999). The sequential hypothesis has even been supported by nap studies: A study by Mednick,

Nakayama, and Stickgold (2003) observed that discrimination thresholds in the same task only improved after a nap containing both NREM and REM sleep but not just NREM. All of these studies suggest that the sequential hypothesis appears more consistent with the literature than the dual process theory.

As with the DPT, there are issues with the sequential hypothesis. Issues regarding the methodology of fragmenting participants’ sleep in order to measure results and that these rely upon correlations between sleep stages and results allow no causal link to be examined. While it may be the more useful model in terms of explaining the role of sleep stages, it is still lacking testability. In an effort to seek a more in-depth view of the process, theories such as the Active System Consolidation (ASC) model instead more towards a more biological focus, focusing on the specific sleep architecture and cortical events that occur during sleep stages and how they may contribute to consolidation.

1.4.1.3 Active System Consolidation Model

(43)

29

qualitative shift in the memory representation as outlined in the CLS model of consolidation (Squire, 1994; McClelland et al., 1995), moving from a veridical based representation to a more semantic, “gist” representation that allows long-term storage. Subsequent periods of REM sleep are thought to act as a period of stabilisation through synaptic consolidation. The important distinction from this and previous models is the focus on the interaction between three aspects of sleep architecture, slow wave activity, hippocampal sharp wave ripples, and sleep spindles (Rasch & Born, 2013). In the following section I will review each of these aspects in more detail and linking them to sleep-based consolidation.

1.4.2 Sleep Architecture

Slow Wave activity

During SWS EEG activity is primarily Slow Wave Activity (SWA). Defined as 0.5-4Hz frequency and <1Hz oscillation, SWA has been demonstrated to play a significant role in consolidation of memory. In animal studies learning new information has led to increases in SWA in subsequent SWS (Kattler, Dijk, & Borbely, 1994; Vyazovskiy, Borbely, & Tobler, 2000), with hemispheric differences observed depending on which paw the rat used during the day (Vyazovskiy, Cirelli, Tononi, & Tobler, 2008). This pattern has been repeated in humans, with Gais, Wagner & Born, (2002) observing increases in SWA during sleep stage SWS after a period of learning word pairs. Furthermore, several studies using transcranial direct current stimulation (tDCS) has provided direct evidence for a causal role of SWA in consolidation. Marshall, Helgadottir, Mölle, & Born (2006) applied tDCS to induce slow oscillating field potentials over the pre-frontal cortex during NREM sleep and observed an increase in SWA that

(44)

30

back only the last decade, these studies demonstrate a direct causal role of SWA and SWS more generally in the consolidation of memory during sleep.

Hippocampal Sharp Wave Ripples

As stated in section 1.1 of the literature review, a key point of the CLS models is the role of sleep in consolidation; sleep is thought to be a period of offline consolidation where repeated activation of hippocampal representation leads to a hippocampal-neocortical transfer. Advances in neuroimaging have revealed a wealth of evidence supporting hippocampal reactivation during sleep in both animal and human studies. Hippocampal SW-R are primarily composed of large amplitude sharp waves in local field potentials and are suggested to be indicators of the most prominent hippocampal activity and are suggested to play an essential role in consolidation. One study that demonstrated their importance was that of Girardeau, Benchenane, Wiener, Buzsaki, and Zugaro (2009) who interfered with hippocampal ripples in sleeping rats who learned a spatial

(45)

31

spindle activity increased further. Furthermore, temporal analyses reveals that spindles and ripples lead to a “spindle-ripple event”, whereby individual ripples become temporally nested in repeated troughs of a spindle (Siapas & Wilson, 1998; Wierzenski, Lubenov Gu, & Siapas, 2009). Taken together, these events may represent the CLS’s suggested transfer of reactivated memories from the hippocampus to neocortical sites (Sirota & Buzaki, 2005; Mölle & Born, 2009).

Sleep Spindles

One of the more recent areas of research in the sleep literature is the role of sleep spindles in memory consolidation. Sleep spindles are characterised as small bursts of activity (11-15Hz) that are relatively fast, typically lasting anywhere between 0.5 – 3s (De Gennaro & Ferrera, 2003). Spindles are most prominent during S2 but are also detectable during SWS, with spindle frequency gradually decreasing as SWS processes (Andrillon et al, 2011; Azumi &

(46)

face-32

name associations (Clemens, Fabo, & Halasz, 2005), and word-pair learning (Gais, Mölle, Helms, & Born, 2002; Schabus et al., 2004; Schmidt et al., 2006; Schabus et al., 2008). One key study in recent years was that of Tamminen, Payne, Stickgold, Wamsley, and Gaskell (2010) who looked at spindles in relation to item integration. Within the study participants learned novel spoken words that had phonological overlap with existing vocabulary (e.g., cathedruke-cathedral) then were either in a wake group or a sleep group that was monitored via polysomnography, with a recall test after the delay and a second recall test one week later. Tamminen et al. observed that participants in the sleep group recalled more words and recognised them faster than the wake group in the 12h delayed test, with the wake group only reporting similar accuracy and speed after the one week delay. Furthermore, recognition of existing vocabulary significantly slowed in both groups after the retention interval, suggesting that novel words had become consolidated and acted as interference for the existing vocabulary. More relevant to the current section, increased spindle activity was strongly correlated with overnight lexical integration of novel words but this was not reflected in increases in recall rate or recognition speed of the novel words, suggesting a central role of spindles in the consolidation and integration of newly learned material into existing cortical networks.

It is also important to note that spindle activity has been linked to region-specific locations for memory consolidation. Previously, Nishida and Walker (2007) demonstrated region-specific spindle activity with relation to a procedural memory task. Participants were trained on a finger tapping task using their left hand and in a subsequent nap period Nishida and Walker observed a significant increase in spindle activity that correlated with performance improvement. When examined further, this correlation was due to spindle activity specifically in the contralateral (right) motor area, with no correlations observed for the LH,

(47)

33

who examined the role of spindles in the integration of novel words into densely or sparsely populated semantic neighbourhood. Within the study participants learned novel words over two sessions and recall was measured immediately after training, the following morning, and one week after. They observed that although recognition amount was equal across both conditions, novel words from the dense semantic neighbourhood were significantly lower in recognition, with Tamminen et al. taking this as evidence for integration of the novel words into the existing semantic network with existing vocabulary acting as interference. They also observed that spindle activity was linked to semantic density, with the sparse semantic neighbourhood related to more sleep spindles during sleep than the dense semantic neighbourhood. Critically, they observed a significant interaction between neighbourhood and hemispheric spindles; the neighbourhood effect observed was specific to the left hemisphere. The result appears consistent with current theories that sleep aids integration of recent experiences into the long-term vocabulary store (Dumay & Gaskell, 2007, 2012) which is suggested to be located in the LH (Ellis, 2004; Ellis et al., 2009). The relevance of hemispheric lateralisation in sleep will be discussed further in Chapter 5.

Yet there is an important point to make about the previous paragraph, the spindle ripple events observed are restricted to fast spindles. Studies have

demonstrated two types of spindles; fast spindles (13-15Hz) are typically distributed over the central and parietal cortex and are typically in S2 whereas slow spindles (10-12Hz) are primarily concentrated around the frontal cortex and are more pronounced in SWS (Anderer et al., 2001; De Gennaro & Ferrara, 2003; Mölle, Bergman, Marshall, & Born, 2011; Terrier & Gottesmann, 1978).

Importantly, the two spindles are also associated with different areas of activity; slow spindles with increased activation in the superior frontal gyrus whereas fast spindles are associated with activation in the hippocampus, suggesting a

(48)

34

differences. Schimdt et al. (2006) observed that compared to a non-learning task spindle activity increased after learning word-pairs, and more specifically, only for slow spindles in frontocentral areas. In the study by van der Helm et al. (2011) within the recognition task they observed that for item memory there was no beneficial effect of sleep, but for context memory, which is hippocampal dependent, there was a significant benefit of sleep, with accuracy significantly correlating with both amount of S2 sleep and frontal fast spindle density. This effect is not limited to declarative memory, with Tamaki et al. (2008, 2009) observing an increase in fast spindle activity following a visuospatial motor task but no such increase for slow spindles. These studies suggest that fast spindles in particular may be vital for neocortical consolidation and links to hippocampal activity.

1.4.3 Sleep’s role in consolidation- Abstraction and Integration

In the last few decades a wealth of literature has demonstrated that sleep is a factor in promotion of associative and a conceptualised form of memory (for a summary, see Chatburn et al., 2014, Mirkovic & Gaskell, 2016), with sleep

benefitting consolidation of numerous forms of abstraction. In a study on relational memory, Ellenbogen Hu, Payne, Titone, and Walker (2007) taught participants a series of premise pairs (A>B, B>C, C>D, D>E, E>F), with an embedded hierarchy (A>B>C>D>E>F). After a delay of 20 minutes, 12 h containing sleep/wake, or 24h participants were examined in an inferential judgement task for novel pairs (e.g., B>D, C>F, B>E). They observed that while the original premise pair recognition was consistent across all groups, only those who had a period of offline delay displayed inferential ability in the novel pair task, suggesting that sleep appears to preferentially facilitate inferential processes through enhancing hierarchical memory. This was expanded upon by Sio,

(49)

35

varying difficulty and after a period of sleep or wake reattempted unsolved

problems. They observed that participants in the sleep group completed a greater number of problems than other groups, but this only applied to problems

categorised as difficult, with no difference in easy problems, a result consistent with existing literature (Cai et al., 2009; Kuriyama, Stickgold, and Walker, 2004). The beneficial role of sleep in abstraction has even been demonstrated in infants, with Gomez, Bootzin, and Nadel (2006) observing that 15-month-old infants who napped between familiarisation and test on an artificial language appeared to remember more abstract relations in the language and generalised to stimuli similar to the familiarisation task. Efforts to create theoretical accounts of

abstraction and integration during sleep have resulted in two prominent theories; the Memory Triage model by Stickgold and Walker (2013) and the information

Overlap to Abstract (iOtA) model by Lewis and Durrant (2011). In understanding

how sleep can lead to abstraction and integration of information, these theories can be broadly split into two key processes: how memory selection occurs during sleep, and how memory reactivation during sleep can affect consolidation.

Memory Selection

In order to avoid consolidating superfluous information consolidation processes during sleep must be capable of preferentially selecting which

Figure

Figure 1.1. Standard model of Complementary Learning Systems. The model assumes two distinct memory systems, a hippocampal-based fast learning store and a neocortical-based slow learning stores, with both occurring in parallel
Figure 1.2. A dog smoking a pipe. Taken from stimuli from Standing
Figure 1.3. Typical sleep stage activity over a single night. Figure taken form
Table 2.1.  Table showing the total participants in each study by culture and ethnicity
+7

References

Related documents

We examine in this section the impact of the network types used for training DeepTMA on its accuracy. As explained in Section IV-B, three different types of networks were

We perform a joint estimation of the physical and risk-neutral dynamics of a matrix affine jump diffusion (MAJD) model in order to study option-implied risks, i.e., risks generated

Kata Kunci: GWPRS, outlier, robust-M, angka kematian ibu Model Geographically Weighted Poisson Regression Semiparametric GWPRS merupakan metode yang dikembangkan dari model GWPR

The accuracy appraisal of thermophysical characteristics determination of optical vitreous silica was carried out by a pulse method by comparison of numerical solution results

The Office of the State Comptroller was honored this past year when it received the Award for Outstanding Achievement in Popular Annual Financial Reporting from the Government

In the study carried out by our group in rats fed an obesogenic diet and treated with resveratrol at a dose of 30 mg/kg body weight/d for 6 weeks, we observed that this

Maximum Equivalent von Mises stresses developed in the neck of the hip prosthesis and maximum total deformation in the implant for static condition were

Mean values of FO peak position (expressed as % of the duration of the accented syllable), the accented syllable duration (on-of) and the distance between the